← sources

Update: Yandex/AliceAI 80B-A3B fine tune progress

source: reddit (reddit:LocalLLaMA)published 2026-09-30 16:30ingested 2026-10-01 02:44

retained text

[loss curve \(taken from the last micro of every step, to explain the variation\)](https://preview.redd.it/yh9zdtvxlosh1.png?width=1579&format=png&auto=webp&s=1ee9257ec5175e2cd08dc54e1c186701dbcd7c6c)

[some help from gemini 3.8 flash high](https://preview.redd.it/iuzw0g2jposh1.png?width=723&format=png&auto=webp&s=e2a83b75fc49f2b2d2d9ea84dfe391d6e7f14b0f)

About 40% of the way done with the initial fine tune. The loss is so spiky because I accidentally used the last loss of each micro, rather than the average of each step

The training live stream is at: [https://figure-bios-expect-cio.trycloudflare.com/](https://figure-bios-expect-cio.trycloudflare.com/) \- and it allows you to inspect any and all of the training data I'm using, if you're interested - I can also provide those roughly 3.5k examples as a dataset. It was generated from sftmill 

Original thread: [https://www.reddit.com/r/LocalLLaMA/comments/1wslgkw/watch\_me\_posttrain\_aliceaifoundation80ba3b\_from/](https://www.reddit.com/r/LocalLLaMA/comments/1wslgkw/watch_me_posttrain_aliceaifoundation80ba3b_from/)

document fetched 2026-10-01 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1wu9ksu/update_yandexaliceai_80ba3b_fine_tune_progress